Update to latest API revision / externalized API client lib

[p3r7]

Relying on an externalized low-level API client lib would allow:

• more projects to rely on it
• have a less monolithic code base
• be more effective when adapting to API changes

Initial idea formulated at #23 (comment).

Would certainly explore using 1Macho/AIDungeonAPI as proposed by #23 (comment).

Subjects to explore:

• have only an async version of API calls (that could be made sync by awaiting them) or propose both (for performance + legibility)
• having a debug option to log queries / responses

[mkualquiera]

One thing to note is that using a library based on asyncio forces the use of asyncio for the project using it, since asyncio is mostly a framework. This is fine for things like Discord or Telegram bots but could not be as great for projects that already work in a different way fundamentally.

[p3r7]

I've made the code more modular (split in several python modules).

On benefit is that it'll ease testing the integration of your lib.

[p3r7]

I haven't tried playing for a while.

It seems that the API changed again 😱.

I'm still using my own "dirty" API client (I haven't found time wrapping my head around python coroutines) but I assume that yours is broke as well, @mkualquiera, right?

[sasha00123]

I guess that's the case, since my bot based on his library is also broken.

[mkualquiera]

From what I could tell, the API is the same... The thing is that they have disabled anonymous access, and now you have to login in order to play.

But I haven't looked into it a lot, I'm not really all that interested in AID anymore. If you guys want I could give you editing access to my repo in case you want to continue working on it.

[p3r7]

I haven’t investigated much but the introduction of the stamina system certainly changed some stuff in the API.

I’ll redo a test ensuring I’m using credentials and not an anonymous account.

I don’t have much time to work on fixing it right now, though. Maybe next week end.

[p3r7]

My code seems to fail with credentials as well.

Given how controversial the new stamina system is I anticipate changes in the weeks to come.

Let's not be too aggressive in trying to keep track with those API evolutions.

[mkualquiera]

I really wish they would just provide an API documentation instead of us having to reverse engineer it every time.
Did the discussion of an official API yield any answers?

[ochen1]

Does anything work atm or should I use a web scraper for my project?

I tried looking at the websocket communications to make my own version but they broke my brain. WAYY to lazy to reverse engineer it.

[p3r7]

Does anything work atm

Nope. Hopefully, if nobody takes the plunge I'll implement it during the Christmas holidays.

I tried looking at the websocket communications to make my own version but they broke my brain.

That's how we did it in the previous iterations of this project.

I can understand, I'm not particularly partial to GQL / Apollo.

should I use a web scraper for my project?

What is your project? How a web scraper could help?

[ochen1]

Sorry for the late reply, the email got drowned out in my email lol

I've already implemented a basic version of AI Dungeon for Python using selenium. Currently, it can log in and create new games with a custom prompt.

My project was to make a general purpose chatbot with GPT-3 through AI Dungeon. However, I've realized that the free model AI Dungeon provides isn't that great for this purpose. I think I'll wait for a public version of GPT-3 before attempting this.

[p3r7]

@ochen1 some peeps seem to have forked this repo to target GPT-3. I don't know if they had a proper early GPT-3 access or if they used AI Dungeon as a backend.

On a related note, this just dropped: https://github.com/AetherDevSecOps/aid_adventure_vulnerability_report

Some takes:

• funny how much of the user base is using AI Dungeon for "18+" adventures. I don't play much but it happened to me once to have a rather explicit answer from the AI even though I'm pretty sure I never activated the NSFW mode.
• they didn't friggin rate-limit the API calls
• I'm not that shocked that they didn't disable introspection in PROD. Malicious ends could brute force their way to the list of endpoints / params anyway. The problem lies more on how open their API is.